Consider a subscription business using a churn prediction model to decide which active subscribers receive a retention offer, typically a discount, ahead of their next billing cycle. Every subscriber has a model-predicted churn risk score, an estimate of a true underlying churn probability the model can only approximate, never observe directly.
Why blanket discounting fails
Offering the discount to every customer, including the large majority who were never actually at meaningful risk of leaving, is expensive for no benefit: the business pays the discount cost on customers it would have kept for free anyway. At any real scale, that cost adds up fast.
Why skipping retention offers entirely fails too
The opposite policy, no targeted offers at all, saves that cost but loses genuinely at-risk, valuable customers who a modest incentive might well have kept. Replacing a lost subscription customer typically costs far more in acquisition spend than the discount that might have retained them would have.
The threshold is where campaign value actually lives
The churn-risk threshold for triggering an offer determines who actually receives it, and campaign value comes down to a direct trade: value from customers retained who would otherwise have churned, minus the cost of offers given to the (often much larger) group who receive one but would have stayed regardless.
Try it yourself
The AI Customer Churn Lab simulates two thousand subscribers with a realistic gap between predicted score and true churn probability, letting you move the threshold and watch offered count, retained count, and net campaign value respond directly.
🧪 Try it yourself: the AI Customer Churn Lab simulation lets you experiment with everything described above directly in your browser.